2014 : Simulation Study to Construct the Prediction Interval for Double Seasonal Holt-Winters

Ir. Mutiah Salamah Chamid


Abstract

This research considers a simulation study for constructing the prediction interval of double seasonal Holt-Winter method. Two approaches have been examined ie percentile error bootstrap and double seasonal block bootstrap. The performance of each method is evaluated with respect to its ability to produce an accurate forecast for several lead times. The performance towards the presence of outlier has also been evaluated. The PE bootstrap is more capable to construct reliable prediction interval for data without outlier. For longer forecasting, accuracy of PE’s prediction interval decreases with increasing lead time, while accuracy of DSB’s prediction interval increases. Interval of PE bootstrap tends to be stable in increasing lead time forecast, whereas interval of DSB is always wider over time.